scsi: target: core: Remove from tmr_list during LUN unlink
authorDmitry Bogdanov <d.bogdanov@yadro.com>
Mon, 18 Oct 2021 13:57:53 +0000 (16:57 +0300)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 18 Nov 2021 18:17:03 +0000 (19:17 +0100)
[ Upstream commit 12b6fcd0ea7f3cb7c3b34668fc678779924123ae ]

Currently TMF commands are removed from de_device.dev_tmf_list at the very
end of se_cmd lifecycle. However, se_lun unlinks from se_cmd upon a command
status (response) being queued in transport layer. This means that LUN and
backend device can be deleted in the meantime and a panic will occur:

target_tmr_work()
cmd->se_tfo->queue_tm_rsp(cmd); // send abort_rsp to a wire
transport_lun_remove_cmd(cmd) // unlink se_cmd from se_lun
- // - // - // -
<<<--- lun remove
<<<--- core backend device remove
- // - // - // -
qlt_handle_abts_completion()
  tfo->free_mcmd()
    transport_generic_free_cmd()
      target_put_sess_cmd()
        core_tmr_release_req() {
          if (dev) { // backend device, can not be null
            spin_lock_irqsave(&dev->se_tmr_lock, flags); //<<<--- CRASH

Call Trace:
NIP [c000000000e1683c] _raw_spin_lock_irqsave+0x2c/0xc0
LR [c00800000e433338] core_tmr_release_req+0x40/0xa0 [target_core_mod]
Call Trace:
(unreliable)
0x0
target_put_sess_cmd+0x2a0/0x370 [target_core_mod]
transport_generic_free_cmd+0x6c/0x1b0 [target_core_mod]
tcm_qla2xxx_complete_mcmd+0x28/0x50 [tcm_qla2xxx]
process_one_work+0x2c4/0x5c0
worker_thread+0x88/0x690

For the iSCSI protocol this is easily reproduced:

 - Send some SCSI sommand

 - Send Abort of that command over iSCSI

 - Remove LUN on target

 - Send next iSCSI command to acknowledge the Abort_Response

 - Target panics

There is no need to keep the command in tmr_list until response completion,
so move the removal from tmr_list from the response completion to the
response queueing when the LUN is unlinked.  Move the removal from state
list too as it is a subject to the same race condition.

Link: https://lore.kernel.org/r/20211018135753.15297-1-d.bogdanov@yadro.com
Fixes: c66ac9db8d4a ("[SCSI] target: Add LIO target core v4.0.0-rc6")
Reviewed-by: Roman Bolshakov <r.bolshakov@yadro.com>
Reviewed-by: Mike Christie <michael.christie@oracle.com>
Signed-off-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/target/target_core_tmr.c
drivers/target/target_core_transport.c

index e7fcbc0..bac1114 100644 (file)
@@ -50,15 +50,6 @@ EXPORT_SYMBOL(core_tmr_alloc_req);
 
 void core_tmr_release_req(struct se_tmr_req *tmr)
 {
-       struct se_device *dev = tmr->tmr_dev;
-       unsigned long flags;
-
-       if (dev) {
-               spin_lock_irqsave(&dev->se_tmr_lock, flags);
-               list_del_init(&tmr->tmr_list);
-               spin_unlock_irqrestore(&dev->se_tmr_lock, flags);
-       }
-
        kfree(tmr);
 }
 
@@ -156,13 +147,6 @@ void core_tmr_abort_task(
                        se_cmd->state_active = false;
                        spin_unlock_irqrestore(&dev->queues[i].lock, flags);
 
-                       /*
-                        * Ensure that this ABORT request is visible to the LU
-                        * RESET code.
-                        */
-                       if (!tmr->tmr_dev)
-                               WARN_ON_ONCE(transport_lookup_tmr_lun(tmr->task_cmd) < 0);
-
                        if (dev->transport->tmr_notify)
                                dev->transport->tmr_notify(dev, TMR_ABORT_TASK,
                                                           &aborted_list);
@@ -234,6 +218,7 @@ static void core_tmr_drain_tmr_list(
                }
 
                list_move_tail(&tmr_p->tmr_list, &drain_tmr_list);
+               tmr_p->tmr_dev = NULL;
        }
        spin_unlock_irqrestore(&dev->se_tmr_lock, flags);
 
index 14c6f2b..e60abd2 100644 (file)
@@ -676,6 +676,21 @@ static void target_remove_from_state_list(struct se_cmd *cmd)
        spin_unlock_irqrestore(&dev->queues[cmd->cpuid].lock, flags);
 }
 
+static void target_remove_from_tmr_list(struct se_cmd *cmd)
+{
+       struct se_device *dev = NULL;
+       unsigned long flags;
+
+       if (cmd->se_cmd_flags & SCF_SCSI_TMR_CDB)
+               dev = cmd->se_tmr_req->tmr_dev;
+
+       if (dev) {
+               spin_lock_irqsave(&dev->se_tmr_lock, flags);
+               if (cmd->se_tmr_req->tmr_dev)
+                       list_del_init(&cmd->se_tmr_req->tmr_list);
+               spin_unlock_irqrestore(&dev->se_tmr_lock, flags);
+       }
+}
 /*
  * This function is called by the target core after the target core has
  * finished processing a SCSI command or SCSI TMF. Both the regular command
@@ -687,13 +702,6 @@ static int transport_cmd_check_stop_to_fabric(struct se_cmd *cmd)
 {
        unsigned long flags;
 
-       target_remove_from_state_list(cmd);
-
-       /*
-        * Clear struct se_cmd->se_lun before the handoff to FE.
-        */
-       cmd->se_lun = NULL;
-
        spin_lock_irqsave(&cmd->t_state_lock, flags);
        /*
         * Determine if frontend context caller is requesting the stopping of
@@ -728,8 +736,16 @@ static void transport_lun_remove_cmd(struct se_cmd *cmd)
        if (!lun)
                return;
 
+       target_remove_from_state_list(cmd);
+       target_remove_from_tmr_list(cmd);
+
        if (cmpxchg(&cmd->lun_ref_active, true, false))
                percpu_ref_put(&lun->lun_ref);
+
+       /*
+        * Clear struct se_cmd->se_lun before the handoff to FE.
+        */
+       cmd->se_lun = NULL;
 }
 
 static void target_complete_failure_work(struct work_struct *work)